Human-computer game war in the offing,
Google: Li Shishi-win possibility for 0-AlphaGo, Li Shishi, go-IT information
On March 8, Google duel AlphaGo Li Shishi press conference following the end of Sina in front of reporters in Korea Seoul Seasons Hotel met Chinese chess national team coach Yu bin at the gate. In cool weather, Yu bin they looked worried, he told us: just get Google News, according to Google engineer's estimate, Li Shishi winning probability is 0.
Competition soon, Li Shishi, and AlphaGo who won and who lost, also was in the cast of a stone. After all, before the 9th playing both, and Fan Hui AlphaGo of professional players and learn from the only one. Two players stand, and make people feel vaguely the mark.
Fan Hui, playing chess and AlphaGo feel like there is a wall in front. Li Shishi, AlphaGo may not like the game before, chipping away at opponents number only themselves and their mangqi under -and that is exactly the same training methods of AlphaGo.
Yu bin's "inside information" some sensational, but Google's Eric Schmidt, Chairman of parent company Alphabet (Eric Schmidt) showed at the press conference are ultra-low profile. Schmidt said that no matter who wins or loses, is actually the human victory --30 years ago, artificial intelligence development experienced a downturn, but now in image recognition, language translation, video testimonials are widely used, even up to and human chess game – he never thought this would become a reality in our lifetime.
Press releases issued, and users in the comments section are happy: watching others leaders speak more in place. We in front of journalists at the press conference after hearing the words behind the music: "find the steps too obvious, there appears to be no confidence. "Who knows what behind DeepMind founder dimisi·hasabisi (Demis Hassabis) get yourself a step:" AlphaGo no instructor, human and machine learning in this regard you will be a huge advantage. ”
But perhaps the biggest step was from Li Shishi. Even though he still believed that human level is stronger than artificial intelligence, but he said that, after you know AlphaGo has a similar intuitive judgment, felt a little nervous . " I'm afraid I beat it 5:0 to a little neighborhood . "On avoiding errors, human machine will always feel the mark.
Compare with caution, SINA journalists two days ahead to reach the two Korea media counterparts, but are strongly on the human side. We reach Korea Shi, take Korea go website Wu Guo Mingao, Director of the network of the car to the hotel. On the highway from Inchon to Seoul, President of Guo smokes, asked us to Ke Jie is true and called "Tesco" chess under the program. When he learned that Ke Jie after verbally expressed interest, even called "boring". He feels that Google's challenge to go brings a high level of attention, any sign of trouble will be the focus.
But when we asked which side he prefers, Guo did not hesitate to choose the Li Shishi, Director: "go Board is too big," as a PhD in computer science, he also participated in a go program development, "computer level to reach this level. "But even so, he said Korea will soon announce its own game artificial intelligence.
Another like-minded peers in the media is Xinhua's Korea correspondent Jin Jingdong. He felt AlphaGo training equivalent to level chess player and chess, actually increase is very small. "Li Shishi's chances even more. "He said
Li Shishi level attention to media go fry pan. Conference questions about Li Shishi, even Korea journalists directly to questions of life and death go circle facing out. "Human represents" a recognized counselling: " artificial intelligence is beyond humanity faces problems of our time, maybe not this time, but it is inevitable in the future . ”
When told Chinese reporters in an interview in the afternoon, hassabis "with the level of discussion" of doubts gives explanations. AlphaGo a total of 18 different versions. When self-discipline, "the parties" is used to choose out of different versions. However, this interpretation seems to be unable to eliminate doubt: different versions of the game at essentially the same level around.
On the day of the dinner, Schmidt repeated once hoped that artificial intelligence would make "more intelligent, better world" vision. SINA in front of journalists also met Yu bin. After a round of drinks, Yu bin faces relax: "write programs as though they did not know who wins or loses. ”
Related reading:
Human-computer game war in the offing: 3 minutes to teach you to read go
人机围棋大战即将上演，
谷歌：李世石赢的可能性为0 - AlphaGo,李世石,围棋 - IT资讯
3月8日中午，谷歌AlphaGo对决李世石的发布会结束之后，新浪前方记者在韩国首尔四季酒店门口碰到了中国围棋国家队教练俞斌。在微凉的天气下，俞斌一行人面色忧虑，他向我们透露：刚刚搞到谷歌内部的消息，据谷歌工程师估算，李世石胜出的可能性为0。
大赛在即，李世石和AlphaGo谁嬴谁输，还是压在众人心上的一块石头。毕竟，9日双方交手之前，和AlphaGo切磋过的专业棋手只有樊麾一个。而两位棋手的表态，都让人隐约觉得心里没谱。
用樊麾的话来说，和AlphaGo下棋感觉面前像有一堵墙。用李世石的话来说，面对AlphaGo备赛不可能像以前那样琢磨对手路数，只能自己和自己下盲棋——这倒是跟AlphaGo的训练方法如出一辙。
俞斌获得的“内幕消息”或许有些耸人听闻，但谷歌母公司Alphabet董事长埃里克·施密特(Eric Schmidt)在新闻发布会上表现出的却是超低姿态。施密特表示，无论谁胜谁负，实际上都是人类的胜利——30年前，人工智能的发展经历过一次低谷，如今却能在图像识别、语言翻译、视频推荐方面被广泛应用，甚至达到了可以和人类对弈围棋的地步——他从来没想过这会在有生之年成为现实。
新闻稿发出来，评论区里的网友都在乐：看人家领导说话多到位。我们前方记者在发布会上听到原话之后也在后面乐：“台阶找得太明显，看来没自信。”谁知道后面DeepMind的创始人迪米斯·哈萨比斯(Demis Hassabis)也给自己找台阶下：“AlphaGo没有导师指导，人类学习在这方面和机器比是个巨大的优势。”
不过最大的台阶或许还是来自李世石。虽然他仍然认为人类的围棋水平还是强于人工智能，但他说，知道AlphaGo具有类似人类直觉的判断能力之后，感到有些紧张。“恐怕我以5比0战胜它有点儿够呛。”在避免失误这件事上，人类面对机器总是觉得心里没谱。
和各方的谨慎态度相比，新浪前方记者这两天接触到的两位韩国媒体同行，倒都是坚定地站在人类这边。我们到达韩国时，搭乘韩国围棋网站乌路网社长郭珉镐的汽车前往酒店。从仁川到首尔的高速公路上，郭社长一边抽烟，一边问我们柯洁是不是真的和叫“什么易购”的程序下了棋。当他得知柯洁只是口头表示有兴趣之后，连呼“没意思”。他觉得，谷歌这次的挑战确实给围棋带来了极高的关注度，任何相关的风吹草动都能成为焦点。
但我们问他更看好哪一方的时候，郭社长毫不犹豫地选择了李世石：“围棋棋盘太大了，”作为一个计算机科学博士，他也参与过围棋程序的开发，“计算机的水平到不了这种程度。”但即便如此，他透露韩国也会很快对外公布自己的围棋人工智能。
另外一个观点相同的媒体同行是新华社的韩国特约记者金敬东。他觉得AlphaGo的自我训练其实相当于和同水平的棋手下棋，实际提升很小。“李世石的胜算更大。”他说
李世石的台阶让关注围棋的媒体炸了锅。发布会关于李世石的提问环节，甚至有韩国记者直接把围棋界是否面临生死存亡的问题抛了出来。“人类代表”有点认怂：“人工智能超越人类是我们这个时代要面临的问题，或许这次不会，但在未来是不可避免的。”
在下午接受中国记者专访的时候，哈萨比斯倒是为“同水平切磋”的怀疑给出了解释。AlphaGo总共有18个不同版本。在自我训练的时候，“双方”使用的是从中选择出来的不同版本。不过这种解释似乎还是无法消除怀疑：不同版本比赛本质上还是在同一个水平打转。
在当天的晚宴上，施密特又重复了一次希望人工智能会让“人类更智能，世界更美好”的愿景。新浪前方记者也再一次见到了俞斌。一巡酒过后，俞斌满脸放松：“好像他们写程序的也不知道谁输谁赢。”
相关阅读：
《人机围棋大战即将上演：3分钟教你看懂围棋》